COMPUTER-BASED AUTOMATED CLASSIFICATION OF PAP SMEAR TESTS USING NEURAL AND FUZZY CLASSIFIERS

摘要

A hierarchical process for automatic classification of Pap smear samples is described and the results are shown. In the preprocessing stage, an edge detection technique is applied to extract contours of cells as well as the nucleus of all cells in sample. In preprocessing stage, we compare the performance of two recently proposed nonlinear filters with a morphological filter used in our technique here. Then, a set of ten features, extracted from each cell, is used to form the feature space. In the next stage, the standard "The Bethesda System" (TBS) rules are translated into fuzzy rules and are used to classify the Pap smear test into "normal" or "abnormal" classes based on the extracted features. In the third stage, a feedforward neural network is applied to the samples for which fuzzy classifier yielded unclear binary classification. The technique mimics the real world practice of cytotechnologist and pathologist in classification of Pap samples, and results to high classification accuracy.